A production-grade machine learning system for weather prediction built entirely in Rust. This project demonstrates the complete ML lifecycle from data collection to model monitoring, using Evcxr Jupyter kernel for interactive exploration.
Auto-updated daily at 06:00 UTC | Last run: Pending first deployment
| City | Country | Current | +24h | +48h | +72h | Rain % | Confidence |
|---|---|---|---|---|---|---|---|
| São Paulo | 🇧🇷 | --°C | --°C | --°C | --°C | --% | -- |
| Rio de Janeiro | 🇧🇷 | --°C | --°C | --°C | --°C | --% | -- |
| São José dos Campos | 🇧🇷 | --°C | --°C | --°C | --°C | --% | -- |
| Campinas | 🇧🇷 | --°C | --°C | --°C | --°C | --% | -- |
| New York | 🇺🇸 | --°C | --°C | --°C | --°C | --% | -- |
| Los Angeles | 🇺🇸 | --°C | --°C | --°C | --°C | --% | -- |
| London | 🇬🇧 | --°C | --°C | --°C | --°C | --% | -- |
| Berlin | 🇩🇪 | --°C | --°C | --°C | --°C | --% | -- |
| Oslo | 🇳🇴 | --°C | --°C | --°C | --°C | --% | -- |
| Tokyo | 🇯🇵 | --°C | --°C | --°C | --°C | --% | -- |
| Shanghai | 🇨🇳 | --°C | --°C | --°C | --°C | --% | -- |
| Chongqing | 🇨🇳 | --°C | --°C | --°C | --°C | --% | -- |
| Nanjing | 🇨🇳 | --°C | --°C | --°C | --°C | --% | -- |
| Dubai | 🇦🇪 | --°C | --°C | --°C | --°C | --% | -- |
| Metric | Rain Prediction | Condition | Temp 24h | Temp 48h | Temp 72h |
|---|---|---|---|---|---|
| Accuracy/RMSE | --% | --% | --°C | --°C | --°C |
| vs Baseline | -- | -- | -- | -- | -- |
- Complete ML Pipeline: Data collection → Preprocessing → Feature Engineering → Model Training → Evaluation → Monitoring
- Multiple ML Libraries: Side-by-side comparison of linfa, smartcore, rustyml, and Burn
- 10 Years of Data: Historical weather data from 2016-2025 for 14 cities worldwide
- 4 Prediction Tasks:
- Rain prediction (binary classification)
- Weather condition classification (6 classes)
- Temperature forecasting (24h, 48h, 72h)
- Multi-target forecasting (temp + humidity + wind)
- Drift Detection: Automated monitoring for data and concept drift
- Live Dashboard: Daily auto-updated predictions via GitHub Actions
- API: Open-Meteo (free, unlimited, no API key required)
- Historical Data: 2016-2025 (10 years)
- Granularity: Hourly observations
- Total Records: ~1.2 million
| Region | Cities |
|---|---|
| Brazil 🇧🇷 | São Paulo, Rio de Janeiro, São José dos Campos, Campinas |
| USA 🇺🇸 | New York, Los Angeles |
| Europe 🇬🇧🇩🇪🇳🇴 | London, Berlin, Oslo |
| Asia 🇯🇵🇨🇳🇦🇪 | Tokyo, Shanghai, Chongqing, Nanjing, Dubai |
RustForMachineLearning/
├── notebooks/ # Jupyter notebooks (Evcxr)
│ ├── 01_data_collection_and_exploration.ipynb
│ ├── 02_preprocessing_and_feature_engineering.ipynb
│ ├── 03_feature_selection_and_model_training.ipynb
│ ├── 04_hyperparameter_tuning.ipynb
│ ├── 05_evaluation_and_validation.ipynb
│ └── 06_drift_detection_and_monitoring.ipynb
├── src/ # Production Rust code
│ ├── lib.rs
│ ├── data/ # Data loading & API client
│ ├── preprocessing/ # Data cleaning & transformation
│ ├── features/ # Feature engineering & selection
│ ├── models/ # ML model implementations
│ ├── training/ # Training utilities
│ ├── evaluation/ # Metrics & visualization
│ ├── monitoring/ # Drift detection
│ └── bin/ # CLI tools
├── data/ # Data storage
│ ├── raw/ # Raw API data
│ ├── processed/ # Cleaned data
│ └── features/ # Engineered features
├── models/ # Trained model artifacts
├── docs/ # Documentation
└── .github/workflows/ # GitHub Actions
- Rust (1.70+)
- Jupyter with Evcxr kernel
- Git
# Clone the repository
git clone https://github.com/yourusername/RustForMachineLearning.git
cd RustForMachineLearning
# Build the project
cargo build --release
# Install Evcxr Jupyter kernel (if not already installed)
cargo install evcxr_jupyter
evcxr_jupyter --install# Start Jupyter
jupyter lab
# Navigate to notebooks/ and open the notebooks in ordercargo run --release --bin daily_predictions┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ DATA │───▶│ PREPROC │───▶│ FEATURES │
│ COLLECTION │ │ & WRANGLING │ │ ENGINEERING │
└─────────────┘ └─────────────┘ └─────────────┘
│ │ │
▼ ▼ ▼
Open-Meteo API Missing values Lag features
14 cities Outliers Rolling stats
10 years Normalization Cyclical encoding
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ FEATURE │───▶│ MODEL │───▶│ TRAINING │
│ SELECTION │ │ COMPARISON │ │ │
└─────────────┘ └─────────────┘ └─────────────┘
│ │ │
▼ ▼ ▼
Correlation linfa Train/Val/Test
Importance smartcore Cross-validation
Recursive elim Burn Early stopping
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ TUNING │───▶│ EVALUATION │───▶│ MONITORING │
│ │ │ │ │ & DRIFT │
└─────────────┘ └─────────────┘ └─────────────┘
│ │ │
▼ ▼ ▼
Grid search Accuracy/F1 Data drift
Random search RMSE/MAE Concept drift
Cross-val ROC/AUC Auto-retrain
| Model | Rain (Acc) | Rain (F1) | Condition (Acc) | Condition (F1) |
|---|---|---|---|---|
| Logistic Regression | -- | -- | -- | -- |
| Decision Tree | -- | -- | -- | -- |
| Random Forest | -- | -- | -- | -- |
| Gradient Boosting | -- | -- | -- | -- |
| Neural Network | -- | -- | -- | -- |
| Ensemble | -- | -- | -- | -- |
| Model | Temp 24h (RMSE) | Temp 48h (RMSE) | Temp 72h (RMSE) |
|---|---|---|---|
| Linear Regression | -- | -- | -- |
| Decision Tree | -- | -- | -- |
| Random Forest | -- | -- | -- |
| Gradient Boosting | -- | -- | -- |
| Neural Network | -- | -- | -- |
| Ensemble | -- | -- | -- |
| Library | Purpose | Status |
|---|---|---|
| linfa | Classical ML | ✅ |
| smartcore | Classical ML | ✅ |
| rustyml | Classical ML | 🔄 |
| Burn | Deep Learning | 🔄 |
- polars: DataFrame operations
- ndarray: N-dimensional arrays
- reqwest: HTTP client
- chrono: Date/time handling
- plotters: Visualization
- statrs: Statistical functions
- Design Document
- API Reference (coming soon)
- Contributing Guide (coming soon)
This project is licensed under the MIT License - see the LICENSE file for details.
- Open-Meteo for providing free weather data
- Rust ML Community for the excellent ML libraries
- Evcxr for the Rust Jupyter kernel